Home // ALLDATA 2026, The Twelfth International Conference on Big Data, Small Data, Linked Data and Open Data // View article

Construction of A Causal Knowledge Graph for Research on Diabetes Comorbidities

Authors:
Xueli Wu
Xuemei Yang
Longchao Wang
Yalan Huang
Xiaoli Tang

Keywords: Diabetes mellitus; Causal knowledge graph; Comorbidity; Causal relationship extraction; Large language model

Abstract:
Diabetes comorbidity is characterized by substantial mechanistic complexity and causal heterogeneity, which correlation-based approaches cannot adequately capture within deep pathological progression pathways. To resolve latent causal structures and mitigate generative hallucination in medical causal mining, this study introduces a hybrid paradigm that integrates physical anchoring, dual-channel evidence awareness, and topological reconstruction, thereby constructing the Diabetes comorbidity Causal Knowledge Graph (Diab- CKG). The framework establishes an atomized corpus indexing coordinate system to ensure traceable extraction and employs Large Language Models under strict ontology constraints to validate prior knowledge and identify novel associations, effectively suppressing generative hallucination. Experimental results demonstrate that the paradigm effectively connects unstructured text with structured reasoning, achieving robust performance in entity recognition and causal extraction, with an end-to-end Strict F1 score of 83.83% and a reduction of the Entity Hallucination Rate to 3.27%. The study delineates a comprehensive causal chain from risk exposure to clinical intervention, providing a logically coherent and computable foundation for modeling comorbidity cascades and supporting clinical decision making.

Pages: 24 to 28

Copyright: Copyright (c) IARIA, 2026

Publication date: May 24, 2026

Published in: conference

ISSN: 2519-8386

ISBN: 978-1-68558-397-2

Location: Venice, Italy

Dates: from May 24, 2026 to May 28, 2026